jobset startup policy implementation

api/jobset/v1alpha2/jobset_types.go

@@ -67,6 +67,11 @@ type JobSetSpec struct {
 	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="Value is immutable"
 	FailurePolicy *FailurePolicy `json:"failurePolicy,omitempty"`
 
+	// StartupPolicy, if set, configures in what order jobs must be started
+	// The targed status of each job is also configurable.
+	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="Value is immutable"
+	StartupPolicy *StartupPolicy `json:"startupPolicy,omitempty"`
+
 	// Suspend suspends all running child Jobs when set to true.
 	Suspend *bool `json:"suspend,omitempty"`
 }
@@ -95,6 +100,7 @@ type ReplicatedJobStatus struct {
 	Succeeded int32  `json:"succeeded"`
 	Failed    int32  `json:"failed"`
 	Active    int32  `json:"active"`
+	Suspended int32  `json:"suspended"`
 }
 
 // +genclient
@@ -177,6 +183,38 @@ type SuccessPolicy struct {
 	TargetReplicatedJobs []string `json:"targetReplicatedJobs,omitempty"`
 }
 
+// Status string
+type Status string
+
+const (
+	// Ready means that the job is ready
+	Ready Status = "Ready"
+
+	// Succeeded means that the job was successful
+	Succeeded Status = "Succeeded"
+)
+
+type StartupPolicy struct {
+	// TargetReplicatedJobStatuses is a list of rules for applying StartupPolicy.
+	// +optional
+	// +listType=map
+	// +listMapKey=name
+	TargetReplicatedJobStatuses []TargetReplicatedJobStartup `json:"targetReplicatedJobStatuses,omitempty"`
+}
+
+type TargetReplicatedJobStartup struct {
+	// Name of ReplicatedJob to apply startup policy
+	Name string `json:"name"`
+
+	// The status you want the ReplicatedJob to be
+	// +kubebuilder:validation:Enum=Ready;Succeeded
+	Status Status `json:"status"`
+
+	// Number of Statuses is how many ReplicatedJob statuses to consider "started"
+	// Empty will mean all status must match the replicas.
+	NumberOfStatuses int32 `json:"numberOfStatuses,omitempty"`
+}
+
 func init() {
 	SchemeBuilder.Register(&JobSet{}, &JobSetList{})
 }

api/jobset/v1alpha2/jobset_webhook.go

@@ -107,6 +107,16 @@ func (js *JobSet) ValidateCreate() (admission.Warnings, error) {
 			allErrs = append(allErrs, fmt.Errorf("invalid replicatedJob name '%s' does not appear in .spec.ReplicatedJobs", rjobName))
 		}
 	}
+	if js.Spec.StartupPolicy != nil {
+		for _, rJobName := range js.Spec.StartupPolicy.TargetReplicatedJobStatuses {
+			if !util.Contains(validReplicatedJobs, rJobName.Name) {
+				allErrs = append(allErrs, fmt.Errorf("invalid replicatedJob name '%s' does not appear in .specReplicatedJobs", rJobName.Name))
+			}
+		}
+		if len(js.Spec.StartupPolicy.TargetReplicatedJobStatuses) > len(validReplicatedJobs) {
+			allErrs = append(allErrs, fmt.Errorf("The length of TargetReplicatedJobStatuses '%d' should not exceed length of valid replicated jobs '%d'", len(js.Spec.StartupPolicy.TargetReplicatedJobStatuses), len(validReplicatedJobs)))
+		}
+	}
 	return nil, errors.Join(allErrs...)
 }
 

api/jobset/v1alpha2/jobset_webhook_test.go

@@ -501,6 +501,71 @@ func TestValidateCreate(t *testing.T) {
 				fmt.Errorf("must be no more than 253 characters"),
 			),
 		},
+		{
+			name: "StartupPolicy Replicated Job does not match any",
+			js: &JobSet{
+				Spec: JobSetSpec{
+					SuccessPolicy: &SuccessPolicy{},
+					ReplicatedJobs: []ReplicatedJob{
+						{
+							Name:     "test-jobset-replicated-job-0",
+							Replicas: 1,
+							Template: batchv1.JobTemplateSpec{
+								Spec: batchv1.JobSpec{},
+							},
+						},
+						{
+							Name:     "test-jobset-replicated-job-1",
+							Replicas: 1,
+							Template: batchv1.JobTemplateSpec{
+								Spec: batchv1.JobSpec{},
+							},
+						},
+					},
+					StartupPolicy: &StartupPolicy{
+						TargetReplicatedJobStatuses: []TargetReplicatedJobStartup{{
+							Name:   "no-exist",
+							Status: Succeeded,
+						}},
+					},
+				},
+			},
+			want: errors.Join(
+				fmt.Errorf("invalid replicatedJob name 'no-exist' does not appear in .specReplicatedJobs"),
+			),
+		},
+		{
+			name: "TargetReplicatedJobStatuses should not be greater length than ReplicatedJobs",
+			js: &JobSet{
+				Spec: JobSetSpec{
+					SuccessPolicy: &SuccessPolicy{},
+					ReplicatedJobs: []ReplicatedJob{
+						{
+							Name:     "test-jobset-replicated-job-0",
+							Replicas: 1,
+							Template: batchv1.JobTemplateSpec{
+								Spec: batchv1.JobSpec{},
+							},
+						},
+					},
+					StartupPolicy: &StartupPolicy{
+						TargetReplicatedJobStatuses: []TargetReplicatedJobStartup{
+							{
+								Name:   "test-jobset-replicated-job-0",
+								Status: Succeeded,
+							},
+							{
+								Name:   "test-jobset-replicated-job-0",
+								Status: Succeeded,
+							},
+						},
+					},
+				},
+			},
+			want: errors.Join(
+				fmt.Errorf("The length of TargetReplicatedJobStatuses '2' should not exceed length of valid replicated jobs '1'"),
+			),
+		},
 	}
 
 	for _, tc := range testCases {